Nvidia, Meta and allies urge targeted rules for open-weight AI models

Nvidia, Meta and Microsoft joined 22 other organisations in warning U.S. policymakers against sweeping curbs on open-weight AI models as competition with China intensifies. The coalition—also including IBM, Palantir, Mistral, Hugging Face, Mozilla, a16z and the Linux Foundation—argues that open-weight models let businesses, researchers and governments download, customise and run models on their own infrastructure, improving control over data, security and compute. They oppose broad restrictions that could weaken American leadership, saying regulators should focus on specific, proven harms such as intellectual-property theft and misuse. The letter backs targeted enforcement rather than blanket bans, and highlights “distillation” (training/improving one model using another model’s outputs) as a common technique for evaluation and validation. The warning comes as the Trump administration considers action against Chinese AI developers accused of using U.S. technology without permission. U.S. Treasury Secretary Scott Bessent said officials are examining whether Chinese models were trained via unauthorised use of outputs, and noted sanctions/Entity List measures could apply if industrial-scale distillation crosses into IP theft. He also stated the administration supports open-source AI and is trying to separate lawful development from alleged copying. Elon Musk publicly supported Nvidia CEO Jensen Huang’s post sharing the letter. Separately, China’s Moonshot AI (Kimi K3) topped a coding benchmark, while U.S. officials have accused Moonshot of distilling from Anthropic’s Fable, keeping the debate active. Crypto-trader angle: the news is an AI policy signal, not a direct crypto regulation, but it may influence risk sentiment around tech/AI infrastructure spending—important for liquidity and correlations—while leaving market impact likely second-order.